The Attacker Has 150 AI Agents. Your SOC Has...
from Security Bros
The numbers don't lie: attackers can operationalize a zero-day exploit in 12 seconds. The average enterprise took 12 weeks to patch Log4Shell. That gap is the whole problem — and it's getting worse. Rocky sits down with Nicolas Popp, a 30-year cybersecurity veteran who built WebObjects at Apple, launched MFA-as-a-service at VeriSign, and now backs the next generation of security companies at Crosspoint Capital. Nico has seen every major wave in this industry. His read on where we are right now: the asymmetry between AI-powered attackers and human-speed defenders has crossed a line that can't be uncrossed. In this episode, they get into the mechanics of what an agentic SOC actually looks like — not the marketing version, the operational one. What does it mean to train AI on 10 years of SOC history? How do you govern agents that are making decisions at machine speed? What happens when your load balancer has a zero-day and there's no patch? And how should MSSPs rethink their entire service model before agents eat them alive? This one gets into the weeds in the best way. If you're building, managing, or investing in security operations, this conversation is required listening. Chapters: 00:00 - Intro: Where's John? Meet Nico Popp 01:45 - 30 years in cyber: from Apple's WebObjects to building MFA-as-a-service 05:10 - The Google AI zero-day moment that changed everything 06:16 - Speed, scale, sophistication: the three vectors of AI-powered attacks 06:34 - Log4Shell took 12 weeks to patch. LLMs generate exploits in 12 seconds. 07:10 - "Joe the Hacker now looks like the Chinese net army" 09:02 - Why the SOC alert backlog problem just became an existential threat 10:57 - Agentic SOC in practice: what the orchestration layer actually does 12:50 - The trust problem: how you start the agentic transformation 13:24 - "We're trading salaries for tokens" — the real budget shift 14:29 - When attackers burn your AI budget on purpose 15:09 - Honeypots never worked. But the digital twin cyber gym might. 16:36 - Vekna and the idea of training your agents in a cyber gym before battle 18:41 - Agent orchestration: the general, the units, and who owns the layer 21:45 - Why operational memory makes agents 10x more valuable over time 23:13 - Advice for legacy vendors: you don't need to own the agentic layer 25:00 - VulnOps: the three-layer framework for closing the patch gap 26:38 - Where Nico would build a startup right now (DLP, CSPM, AppSec) 28:49 - AI, wealth concentration, and the social question nobody wants to answer 30:28 - The one-person unicorn: how many years until it happens? 33:13 - Identity Action Management: the governance problem nobody has solved 36:38 - Why the "governance harness" is replacing "sandbox" as the right frame 38:09 - Agentic VulnOps in action: patching, control reconfiguration, detection rules 43:59 - Third-party and legacy software: the attack surface you can't fix 46:22 - Zest Security: AI that generates the WAF rule and the Terraform code 49:12 - Slack as the audit trail for what your agents are doing 51:00 - How professional services firms need to become software companies 54:50 - Wrap-up and where to follow Nico on LinkedIn
Embed this episode
No transcript yet.